1. If these were your hypotheses then ome-tailed tests seem appropriate

2. Check out https://www.researchgate.net/publication/329398465_Examining_the_Testing_Effect_in_University_Teaching_Retrievability_and_Question_Format_Matter?_sg=84nhmA8ZUqnbRSAwFSFQTKTJpAmyHi57qb8JDgTiVe6gvSB5eQye_RPo0kP_lcCRSDOgpI2TbkdqhP-Gxazt5trV8Vy1nhy9BXUajUqv.Ks4Rb9-KJ5NyKDwxQXJRrk1fPvQupO0DCaAEZ14_0KFJ1vh_d23FBHEMu6jhxcTNnCYgLL2ZMdkr8E0zBH1B4g

1. If these were your hypotheses then ome-tailed tests seem appropriate

2. Check out https://www.researchgate.net/publication/329398465_Examining_the_Testing_Effect_in_University_Teaching_Retrievability_and_Question_Format_Matter?_sg=84nhmA8ZUqnbRSAwFSFQTKTJpAmyHi57qb8JDgTiVe6gvSB5eQye_RPo0kP_lcCRSDOgpI2TbkdqhP-Gxazt5trV8Vy1nhy9BXUajUqv.Ks4Rb9-KJ5NyKDwxQXJRrk1fPvQupO0DCaAEZ14_0KFJ1vh_d23FBHEMu6jhxcTNnCYgLL2ZMdkr8E0zBH1B4g

Hello Lisa,

If your hypothesis was directional, then I agree with Sven that halving the nondirectional p-value would give you the matching probability under the null hypothesis. You'd have to double the error rate for the CI in order to get an analog, one-sided limit estimate (e.g., instead of looking at 95% CI, look at 90% CI, one side only). As you appear to be using SPSS, just click the "options" button in the LR dialog box and adjust the CI for exp(B) to the desired value.

Good luck with your work.

This seems like an ideal case for Bayesian methods. You can incorporate your prior knowledge about directionality.

@Sven Greving, thank you for the link to your paper. It is very useful!

David Morse thank you for your clear explanation! I assume adapting the 95% CI to 90% CI is also the case for boostrapped CI?

Another question also came to mind. Are you allowed to create an one-tailed hypothesis but to examine the differences in both way? For example, if literature points towards one direction but you are curious if the effect also goes the other way, can you then create an one-sided hypothesis and examine it in the two-way manner? The literature "forces" you to go one way with your hypothesis, but you actually want to measure it both ways. Is there some way that allows you to create a one-directed hypothesis, but to explain the results from two-sided. For example, is it allowed to explain why an one-directed hypothesis is formulated but why the results are interpreted two-ways?

It sounds to me like you ask the following: If we did not find evidence for a directional hypothesis can we proceed testing a undirectional hypothesis or a hypothsis directed in the opposite way ?

Keep in mind that when testing your first hypothesis you already "spent" a 5 percent probability of reporting an effect that is not there (alpha error). Stacking additional hypothesis tests would mean to further increase the alpha error. This would not be a problem if you knew beforehand that you would test several hypotheses because then you could adjust alpha levels. However, the crux is that with decreased alpha levels, effects need to be bigger in order to be detected as significant. When you thus not find an effect that has been expected to be there because prior works said it should be there, finding the opposite effect (that is large enough to be detected despite alpha level correction) would at least raise an eyebrow of the community.

Always assuming alpha levels of 5percent until you found a significant result would be downright fraud.

To circumvent problems like this you could always turn towards Bayesian analyses as already suggested because they allow for testing of hypotheses on the basis of data (unlike frequentist approaches that test data on the basis of hypotheses). You could thus compare two opposite hypotheses and an undirectional hypothesis while incorporating knowledge from prior studies.